import gradio as gr
import pandas as pd
import json
from pathlib import Path

# Load data
def load_data():
    tables = Path("artifacts/tables")
    figures = Path("artifacts/figures")
    
    app_summary = pd.read_csv(tables / "app_summary.csv")
    pricing = pd.read_csv(tables / "pricing_decisions.csv")
    sentiment = pd.read_csv(tables / "sentiment_per_app.csv")
    
    with open(figures / "kpis.json") as f:
        kpis = json.load(f)
    
    return app_summary, pricing, sentiment, kpis

def get_overview():
    app_summary, pricing, sentiment, kpis = load_data()
    return (
        f"Total Apps Analyzed: {kpis['total_apps']}",
        f"Real Reviews Scraped: {kpis['total_real_reviews']}",
        f"Synthetic Reviews: {kpis['total_synthetic_reviews']}",
        f"Months Analyzed: {kpis['months_analyzed']}",
        f"Best Satisfaction: {kpis['best_app_satisfaction']}",
        f"Highest Churn Risk: {kpis['highest_churn_app']}",
        f"Lowest Churn Risk: {kpis['lowest_churn_app']}"
    )

with gr.Blocks(title="Productivity App Analyzer", theme=gr.themes.Soft()) as demo:
    
    gr.Markdown("""
    # Productivity App Analyzer
    ### AI for Big Data Management — ESCP Business School
    *Analyzing TickTick, Todoist, Microsoft To-Do, and Any.do using real reviews + synthetic data*
    """)

    # KPI Row
    with gr.Row():
        kpi1 = gr.Textbox(label="Apps Analyzed", interactive=False)
        kpi2 = gr.Textbox(label="Real Reviews", interactive=False)
        kpi3 = gr.Textbox(label="Synthetic Reviews", interactive=False)
        kpi4 = gr.Textbox(label="Months Analyzed", interactive=False)

    with gr.Row():
        kpi5 = gr.Textbox(label="Best Satisfaction", interactive=False)
        kpi6 = gr.Textbox(label="Highest Churn", interactive=False)
        kpi7 = gr.Textbox(label="Lowest Churn", interactive=False)

    # Charts
    gr.Markdown("## Visualizations")
    with gr.Tabs():
        with gr.Tab("Sentiment Analysis"):
            gr.Image("artifacts/figures/sentiment_per_app.png",
                     label="Sentiment Distribution per App")

        with gr.Tab("Churn & Satisfaction"):
            gr.Image("artifacts/figures/churn_satisfaction_per_app.png",
                     label="Churn Risk & Satisfaction Scores")

        with gr.Tab("Usage Trends"):
            gr.Image("artifacts/figures/monthly_usage_trends.png",
                     label="Monthly Active Users - 18 Months")

        with gr.Tab("ARIMA Forecasts"):
            gr.Image("artifacts/figures/arima_forecasts.png",
                     label="6-Month User Forecasts per App")

    # Pricing Decisions
    gr.Markdown("## Pricing Recommendations")
    
    def get_pricing_table():
        _, pricing, _, _ = load_data()
        return pricing[["app_name", "positive_ratio",
                         "negative_ratio", "avg_churn",
                         "avg_satisfaction", "pricing_action"]].round(3)

    pricing_table = gr.Dataframe(
        label="Rule-Based Pricing Decisions",
        interactive=False
    )

    # App Summary
    gr.Markdown("## App Performance Summary")
    
    summary_table = gr.Dataframe(
        label="Average Churn, Satisfaction & Daily Tasks",
        interactive=False
    )

    # Load on startup
    def load_all():
        app_summary, pricing, sentiment, kpis = load_data()
        k = get_overview()
        pricing_df = pricing[["app_name", "positive_ratio",
                               "negative_ratio", "avg_churn",
                               "avg_satisfaction", "pricing_action"]].round(3)
        return (*k, pricing_df, app_summary)

    demo.load(
        fn=load_all,
        outputs=[kpi1, kpi2, kpi3, kpi4, kpi5, kpi6, kpi7,
                 pricing_table, summary_table]
    )

    gr.Markdown("""
    ---
    *Data: 400 real Google Play reviews + 4,000 synthetic reviews | 
    Methods: VADER Sentiment, ARIMA Forecasting, Rule-Based Pricing*
    """)

demo.launch()